Volcano Engine released the Doubao Model 1.8 at the 2025 Winter Force Prime Power Conference. Tan Dai, president of Volcano Engine, introduced that the new version has been enhanced for agent tasks, while improving multimodal understanding capabilities and providing more flexible context management to adapt to longer and more complex business processes and interaction scenarios.
The conference simultaneously disclosed the operating data, saying that as of December this year, the average daily call volume of the bean bag model has exceeded 50 trillion yuan, an increase of more than ten times from December last year, and an increase of about 417 times compared with the initial release of the bean bag. The relevant data caliber is mostly expressed as "call volume/usage" in public reports, and whether it is counted by token, number of requests or other indicators has not yet been more detailed and unified.
At the industry level, the increase in the average daily call scale usually means that the model is "normalized" embedded in more business links, but it also amplifies risks such as content compliance, data security, and model hallucinations, and puts forward higher requirements for pre-launch evaluation, permission control, and audit traces.
FAQ
Q: What version is the Doubao model 1.8 update?
A: The latest iterative version of the Doubao model focuses on strengthening Agent capabilities, multi-modal understanding and context management.
Q: Who released Doubao 1.8 at the Force Momentum Conference?
A: Tan Dai, President of Volcano Engine, released and introduced the main upgrade points at the conference.
Q: What does the "average daily call volume of more than 50 trillion yuan" refer to in the Doubao model?
A: Public reports mostly use "calls/usage" to summarize and disclose, specifically by token, number of requests, or other calibers, and the report does not give a unified and detailed definition.
Q: What does the core capability upgrade of Doubao 1.8 mean for enterprises?
A: It is more conducive to building executable agent processes, processing multimodal inputs such as graphics and text, and improving controllability and usability in long-context tasks.
Q: What are the most common risk points when enterprises access large models?
A: Billing caliber and cost fluctuations, data compliance and permission management, output hallucinations and security audits, and copyright and inappropriate content risks of multimodal content.
